Conserved energy for v-rescale thermostat drifts

GROMACS version: 2019.4, 2020.6
GROMACS modification: No


I was running NVT equilibrations for a small protein (about 100 residues) in TIP3P water (about 30,000 atoms in total) using the v-rescale thermostat (varying tau_t) with standard simulation parameters and the AMBER99SB-ildn force field as provided in GROMACS:

integrator = md
nsteps = 200000
dt = 0.001

constraints = h_bonds

cutoff_scheme = verlet
coulombtype = PME
fourierspacing = 0.12
pme_order = 6

tcoupl = v_rescale
tc_grps = Protein non-Protein
tau_t = 0.1 0.1
ref_t = 300 300

gen_vel = yes
gen_temp = 300
continuation = no

Before the equilibration, the system was well minimised (first steepest decent, then conjugate gradient up to the single-precision limit).

Now, when I check the temperature evolution, everything seems fine (see figure below).

But when I look at the “Conserved Energy” quantity written to the EDR file (which as I understand refers to the conserved quantity for the thermostat as described here: J. Chem. Phys. 126, 014101 (2007);, it shows an apparent drift. The absolute drift has a small scale but compared to the fluctuation of this quantity it still seems substantial.

My questions are: Should I worry about this? If yes, what could potentially cause this?

I already tested with “DispCorr = No” (as I guess is recommended for the force field) and “DispCorr = EnerPress” (not shown) but the trend is the same.

Numerical integration never perfectly conserves energy. This drift is very small, so nothing to worry about.
The largest sources of drift are usually the constraint algorithm(s), which always give a negative drift, and the errors allowed by the verlet-buffer-tolerance parameter, which usually give a positive drift.

1 Like